В статье рассматривается подход к исследованию структурных и фазовых изменений углеродных наноструктурированных материалов на основе интеграции методов электронной спектроскопии и технологий искусственного интеллекта. Анализируются возможности определения электронной и поверхностной структуры материалов с использованием рамановской спектроскопии и рентгеновской фотоэлектронной спектроскопии (XPS). Предлагается концептуальная модель обработки спектральных данных с применением алгоритмов машинного обучения, классификации фаз и оптимизации структурных параметров. Данный подход создаёт перспективные возможности для выявления взаимосвязи «структура–свойство» углеродных наноматериалов и целенаправленного управления параметрами их синтеза.

Ключевые слова: углеродные наноматериалы, наноструктура, фазовые превращения, электронная спектроскопия, рамановская спектроскопия, XPS, искусственный интеллект, машинное обучение, структурные параметры, оптимизация.

Carbon is one of the most abundant and structurally diverse elements in nature. Its atoms can form materials with fundamentally different properties through various hybridization states. Although graphite, diamond, graphene, fullerenes, carbon nanotubes, nanodiamond, amorphous carbon, and other carbon allotropes consist of the same chemical element, their physical, chemical, mechanical, and electronic properties differ significantly. These differences primarily result from variations in atomic arrangement, bonding configurations, and electronic structure. The development of nanotechnology has expanded the application of carbon-based materials in next-generation electronic devices, sensors, energy-storage systems, composites, catalytic systems, and other high-technology fields. At the same time, the final properties of nanomaterials are highly sensitive to synthesis conditions. Even small variations in temperature, pressure, precursor composition, processing time, chemical environment, and other factors can affect their structural state.

Spectroscopic methods are particularly important for investigating carbon nanostructures. Raman spectroscopy provides valuable information about structural disorder, defects, graphitization degree, and hybridization states in carbon materials. In modern research, Raman spectroscopy is widely used to characterize the microstructural properties of graphene, nanotubes, carbon fibers, amorphous carbon, and related materials.

X-ray photoelectron spectroscopy (XPS) is primarily used to investigate the chemical composition of material surfaces, the chemical states of elements, and bonding configurations. As a surface-sensitive technique, XPS enables the analysis of the chemical states present in the outermost layers of carbon nanostructures. One of the main limitations of conventional spectroscopic analysis is the difficulty of rapidly and comprehensively processing large volumes of spectral data manually. Data interpretation becomes particularly challenging when different phases exhibit similar spectral characteristics or when a material consists of a mixture of several structural states. In this context, artificial intelligence and machine learning technologies provide new opportunities in materials science. Recent studies have demonstrated the application of machine learning to materials design, property prediction, spectroscopic data classification, and optimization of synthesis parameters. The main objective of this study is to develop a comprehensive methodological approach for identifying phase and structural transformations in carbon nanostructured materials based on electronic spectroscopic data and for processing these data using artificial intelligence to determine optimal structural parameters. The research methodology consists of three main stages:

1. Acquisition of experimental spectroscopic data from carbon nanostructured materials;

2. Extraction of structural and electronic descriptors from the spectral data;

3. Determination of the phase state and prediction of optimal parameters using machine learning.

In the first stage, Raman spectra and XPS spectra of the materials are obtained, along with XRD, electron microscopy, or other instrumental analysis results when necessary. In Raman spectroscopy, the intensity, intensity ratio, peak position, full width at half maximum, and spectral background of the D and G bands are considered important structural descriptors. Recent reviews have demonstrated the relationship of the D and G bands with defects, structural disorder, and graphitization degree in carbon materials. In the second stage, digital processing of the spectra is performed. Raw spectral data undergo noise reduction, baseline correction, normalization, and, where necessary, spectral deconvolution. This stage helps reduce the risk of the artificial intelligence model learning spurious correlations. For XPS data, the principal photoelectron signals of C 1s, O 1s, and other available elements are analyzed. Deconvolution of the C 1s spectrum into individual components provides information about different chemical bonding states of carbon. The main advantage of XPS is its ability to determine surface chemistry and the chemical states of elements. In the third stage, the extracted parameters are introduced into a machine learning model. The input parameters may include:

– Raman D/G intensity ratio;

– positions of the G and D bands;

– spectral band widths;

– relative proportions of XPS components;

– sp²/sp³ ratios;

– proportion of oxygen-containing functional groups;

– degree of crystallinity;

– synthesis temperature;

– pressure;

– processing time;

– precursor composition.

The output parameters may include the phase state of the material, degree of structural ordering, defect concentration, or targeted physical properties.

The properties of carbon materials depend on their sp² and sp³ hybridization states, degree of crystallinity, surface chemistry, and defect concentration. In nanostructures, the increased proportion of surface atoms enhances the influence of adsorption, functional groups, and defects on the material’s properties. XPS enables the determination of surface chemical composition and electronic states, while Raman spectroscopy provides information about the vibrational and structural characteristics of the material. The G and D bands in Raman spectra are associated with sp²-bonded carbon and structural disorder and defects, respectively. Although the D/G ratio is useful for evaluating the structural state, its application as a single universal criterion is limited. Phase transformations in carbon materials can occur as a result of thermal, mechanical, pressure-induced, or chemical effects. Raman spectroscopy enables rapid and non-destructive monitoring of these transformations, whereas XPS provides information about their chemical aspects, including oxidation state, functional groups, and bonding configurations. Therefore, the integrated analysis of Raman and XPS data enables a more reliable assessment of the structural and chemical states of the material. This multimodal approach provides a basis for the comprehensive analysis of spectral descriptors using artificial intelligence algorithms.

This workflow enables the conventional “synthesis–measurement–analysis” cycle in materials science to be transformed into a data-driven and controllable process. A key advantage of artificial intelligence in materials science is its ability to identify nonlinear relationships within complex and multidimensional datasets. The relationship between synthesis conditions and structural characteristics of a material often cannot be adequately described by a simple linear equation. In supervised machine learning, a model is trained using data associated with predefined outcomes. Based on Raman and XPS spectra, material samples can be classified into specific structural classes.

Fig. 1.Morphological and structural characteristics

Fig. 2. Electronic and electronic-structure characteristics

When a model receives a new spectrum, it can predict which structural group it most closely resembles. In supervised learning, several algorithms can be applied. For spectral data, neural networks, particularly convolutional neural networks (CNNs), are of considerable interest because of their ability to automatically extract local features from spectra. A 2024 study applied deep learning to the rapid identification of two-dimensional materials using Raman spectra, combining spectral data augmentation with classification. However, high predictive accuracy alone does not necessarily indicate that a model is scientifically reliable.

Overfitting may occur when working with small datasets. In materials science, limited data availability, data quality, and external validation are considered important challenges. Therefore, the proposed methodology should divide the dataset into training, validation, and test subsets. In addition to overall accuracy, the model should be evaluated using precision, recall, F1-score, confusion matrix, and performance on external test samples. In material optimization, the key objective is not only to classify existing samples but also to determine which technological parameters should be selected to obtain a material with the desired properties. The model analyzes the effects of these parameters on the structural characteristics of the material. In the next stage, Bayesian optimization or active learning can be used to identify the most promising parameter combinations. The main advantage of active learning is that, rather than selecting each new experiment randomly, it enables the selection of experiments that provide the greatest amount of information for improving the model.

The integration of electronic spectroscopy and artificial intelligence provides opportunities to address several important challenges in the investigation of carbon nanomaterials. First, the high dimensionality of spectroscopic data enables the detection of subtle spectral differences that may not be readily identified through direct human analysis. Second, data obtained from multiple spectroscopic techniques can be integrated into a single analytical model. However, multimodal data integration presents a standardization challenge. Raman and XPS data differ in measurement units, signal ranges, noise characteristics, and physical meaning. Therefore, before combining them into a single dataset, the data must be normalized and transformed into compatible descriptors. Another important issue is model interpretability. A scientific model in materials science should not be limited to answering the question “Which phase is present?” It should also, where possible, address questions such as “Why was this phase predicted?”, “Which spectral features had the greatest influence on the decision?”, and “Which technological parameter caused the most significant structural change?” Therefore, Explainable AI approaches and interpretation tools such as feature importance and SHAP represent promising directions.

Experiments involving nanomaterials are often expensive, time-consuming, and limited in terms of reproducibility. Consequently, constructing large, high-quality datasets can be challenging. For this reason, algorithms capable of working with small datasets, transfer learning, active learning, and models enriched with physical knowledge are particularly important. Integrating physical laws into machine learning models allows AI to account not only for statistical relationships but also for known physical constraints of materials science. This approach can improve the generalization capability of the model and help prevent physically implausible predictions. The scientific significance of the proposed methodology lies in its ability to consider the structural, chemical, and electronic properties of carbon nanostructured materials within a unified data framework. From a practical perspective, this approach can be used to:

– rapidly classify the structural state of carbon nanomaterials;

– identify phase transformations through spectral features;

– evaluate the chemical state of the surface;

– determine parameters associated with defects and structural disorder;

– model the effects of synthesis parameters on material structure;

– predict optimal technological conditions;

– intelligently select new experiments;

– implement data-driven optimization aimed at reducing the number of experiments.

The application of artificial intelligence to nanomaterial design and process optimization increasingly encompasses these types of tasks.

The study of phase and structural transformations in carbon nanostructured materials is one of the important areas of modern materials science. The functional properties of such materials depend on numerous interrelated factors, including interatomic bonding, hybridization, crystallization, defects, surface chemistry, and electronic structure. Raman spectroscopy serves as an effective tool for characterizing the vibrational and structural properties of carbon nanostructures, whereas X-ray photoelectron spectroscopy (XPS) is widely used to investigate the elemental composition and chemical state of their surfaces. The combined application of these methods enables a comprehensive characterization of the structural and chemical states of the material.

Artificial intelligence, in turn, provides opportunities for processing large volumes of spectroscopic data, automatically extracting spectral features, classifying material phases, and predicting structural parameters. The main advantage of machine learning lies not only in achieving high predictive accuracy but also in its ability to identify complex multidimensional relationships between processing conditions, structure, and properties. Within the proposed research concept, the integration of Raman and XPS data into a unified multimodal dataset, classification of structural states using machine learning, and selection of subsequent experimental points through Bayesian optimization or active learning are proposed. Such an approach makes it possible to transform the conventional synthesis–analysis–resynthesis cycle into a data-driven adaptive optimization system.

Independent experimental validation of the results generated by artificial intelligence models, ensuring dataset quality, preventing overfitting, and implementing model interpretability are essential requirements. In modern materials science, the effectiveness of AI is strongly dependent on data quality and external validation of the developed models. The integration of electronic spectroscopy and artificial intelligence provides a scientific basis for transforming the investigation of carbon nanostructured materials from a primarily descriptive approach toward a predictive and optimization-oriented methodology. In the future, this methodology can contribute to the identification of new structural modifications of carbon nanomaterials, targeted control of their desired properties, and data-driven selection of optimal synthesis parameters.

References:

1. Содиков Ф. Х. У. и др. Исследование структурного анализа и свойств бронзового элемента //Молодой ученый Учредители: ООО» Издательство Молодой ученый». — №. 16. — С. 58–61.
2. Yuan R., Guo Y., Gurgan I. et al. Raman spectroscopy analysis of disordered and amorphous carbon materials: A review of empirical correlations // Carbon. — 2025. — Vol. 238. — Article 120214.
3. Childres I., Jauregui L. A., Tian J., Chen Y. P. Raman Spectroscopy of Graphene and Related Materials // New Developments in Photon and Materials Research. — 2013.
4. Ferrari A. C., Basko D. M. Raman spectroscopy as a versatile tool for studying the properties of graphene // Nature Nanotechnology. — 2013. — Vol. 8. — P. 235–246.
5. Nanda Gopala Krishna D., Philip J. Review on surface-characterization applications of X-ray photoelectron spectroscopy (XPS): Recent developments and challenges // Applied Surface Science Advances. — 2022. — Vol. 12. — Article 100332.
6. Moulder J. F., Stickle W. F., Sobol P. E., Bomben K. D. Handbook of X-ray Photoelectron Spectroscopy. — Physical Electronics, 1992.
7. Woodruff D. P., Delchar T. A. Modern Techniques of Surface Science. — Cambridge University Press, 1994.
8. O’Connor H. M., et al. Characterization of Carbon Nanostructures by Photoelectron Spectroscopies // Materials. — 2022. — Vol. 15. — Article 4434.
9. Larrude D. G., et al. Raman spectroscopy of carbon materials and their composites: Graphene, nanotubes and fibres // Progress in Materials Science. — 2023. — Vol. 135. — Article 101089. (
10. Papadimitriou I., Gialampoukidis I., Vrochidis S., Kompatsiaris I. AI methods in materials design, discovery and manufacturing: A review // Computational Materials Science. — 2024. — Vol. 235. — Article 112793. (
11. Machine learning applications in nanomaterials: Recent advances and future perspectives // Chemical Engineering Journal. — 2024.